A coarse aggregate UHPC wet joint shrinkage stress real-time prediction method and system

Abstract

This invention discloses a real-time prediction method and system for shrinkage stress in wet joints of coarse aggregate UHPC, specifically relating to the field of data processing and predictive analysis technology based on graph neural networks. It addresses the problem of time delays in existing wet joint shrinkage stress prediction, which prevents timely early warning. The method involves real-time acquisition of stress, temperature, and humidity data; identification of temperature abrupt events and calculation of stress response lag time; establishment of a stress-temperature change rate correlation model to determine nonlinear change stages; construction of a spatiotemporal map of monitoring points and analysis of spatial dependencies using graph neural networks; dynamic identification of risk convergence nodes; prediction of shrinkage stress extrema and their occurrence time based on stress data from risk nodes; and generation and transmission of maintenance control commands to field equipment when the predicted value exceeds the material's tensile strength threshold. This achieves real-time, accurate prediction of shrinkage stress and proactive maintenance intervention.

Description

Technical Field

[0001] This invention relates to the field of data processing and predictive analysis technology of graph neural networks, and more specifically, to a method and system for real-time prediction of shrinkage stress in wet joints of coarse aggregate UHPC. Background Technology

[0002] In the construction of ultra-wide composite girder cable-stayed bridges, coarse aggregate ultra-high performance concrete is widely used in the wet joints of bridge decks due to its high strength and durability. During the hardening process, stress concentration easily occurs in wet joints due to material shrinkage and environmental factors. Improper control can lead to early structural cracking, affecting the overall safety of the bridge. To prevent such problems, dynamic prediction of shrinkage stress in wet joints is necessary to guide on-site curing measures and construction adjustments.

[0003] Existing prediction methods rely on post-processing and analysis of monitoring data. The calculation process requires accumulating data over a certain period before being performed, resulting in a significant delay between the predicted results and the actual stress evolution. This delay prevents on-site personnel from obtaining timely early warning information during critical stages of accelerated shrinkage stress growth, making it difficult to effectively intervene in the generation and development of cracks. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, the present invention provides a method and system for real-time prediction of shrinkage stress in wet joints of coarse aggregate UHPC to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A method for real-time prediction of shrinkage stress in wet joints of coarse aggregate UHPC includes the following steps:

[0007] S1. Real-time acquisition of stress, temperature and humidity data at wet joint monitoring points;

[0008] S2. Based on temperature data, identify temperature abrupt events where the rate of change in ambient temperature exceeds the threshold of normal construction fluctuations.

[0009] S3. For temperature mutation events, analyze the stress data within a preset time window after the event occurs, and calculate the response lag time of the stress change rate relative to the temperature mutation event.

[0010] S4. Based on stress data, temperature data, and response lag time, analyze the correlation characteristics between stress change rate and temperature change rate. When the ratio of stress change rate to temperature change rate exceeds a set threshold, it is determined that the stress nonlinear change stage has been entered.

[0011] S5. During the nonlinear stress change stage, a spatiotemporal map is constructed based on stress data, temperature data and humidity data of all monitoring points. A graph neural network is used to analyze the spatial dependence between nodes, and risk convergence nodes are identified through dynamic node centrality calculation.

[0012] S6. Determine the stress data of the risk convergence node as the core prediction object, predict the extreme value of shrinkage stress and its occurrence time within a set period in the future, and generate maintenance control instructions for the risk convergence node and output them to the on-site maintenance equipment.

[0013] Furthermore, real-time acquisition of stress, temperature, and humidity data at wet joint monitoring points includes:

[0014] Stress sensors, temperature sensors, and humidity sensors are installed on the surface and in key areas inside the wet joint.

[0015] Stress data is continuously collected using a stress sensor, temperature data is continuously collected using a temperature sensor, and humidity data is continuously collected using a humidity sensor.

[0016] The collected stress, temperature, and humidity data are simultaneously transmitted to the data storage location.

[0017] Furthermore, based on temperature data, temperature abrupt events where the rate of change in ambient temperature exceeds the threshold for typical construction fluctuations are identified, including:

[0018] Establish a time series of temperature data based on real-time acquired temperature data;

[0019] Calculate the rate of temperature change between adjacent time points in a time series of temperature data;

[0020] The threshold for fluctuations in conventional construction conditions is obtained by statistically analyzing historical construction environment temperature data collected in advance.

[0021] The calculated rate of temperature change is compared in real time with the conventional construction fluctuation threshold.

[0022] When the rate of temperature change continuously exceeds the normal construction fluctuation threshold, it is identified as a temperature abrupt event.

[0023] Furthermore, for temperature abrupt events, stress data within a preset time window after their occurrence is analyzed, and the response lag time of the stress change rate relative to the temperature abrupt event is calculated, including:

[0024] Determine a preset time window starting from the moment the temperature abrupt change event occurs;

[0025] Extract stress data and calculate the rate of stress change within a preset time window;

[0026] Analyze the trend of stress change rate within a preset time window to identify the inflection point from the initial response to a significant acceleration in stress change rate;

[0027] Calculate the time difference between the occurrence of the temperature abrupt change event and the occurrence of the inflection point of the stress change rate, and use this time difference as the response lag time of the stress change rate relative to the temperature abrupt change event.

[0028] Furthermore, based on stress data, temperature data, and response hysteresis time, the correlation characteristics between the stress change rate and the temperature change rate are analyzed. When the ratio of the stress change rate to the temperature change rate exceeds a set threshold, it is determined that the stress has entered a nonlinear change stage, including:

[0029] The corresponding analysis period for stress data and temperature data is determined based on the response lag time.

[0030] Calculate the rate of stress change and the rate of temperature change within the corresponding analysis period;

[0031] Calculate the real-time ratio of the rate of stress change to the rate of temperature change;

[0032] The stress-temperature coupling response threshold is set based on the material properties of coarse aggregate UHPC.

[0033] When the real-time ratio of the rate of stress change to the rate of temperature change continuously exceeds the stress-temperature coupling response threshold, it is determined that the stress has entered the nonlinear change stage.

[0034] Furthermore, during the nonlinear stress variation stage, a spatiotemporal map is constructed based on stress, temperature, and humidity data from all monitoring points. A graph neural network is used to analyze the spatial dependencies between nodes, and risk convergence nodes are identified through dynamic node centrality calculations, including:

[0035] Each monitoring point is used as a graph node, and the spatial distance between monitoring points is used as an edge to construct the graph structure;

[0036] The stress data, temperature data, and humidity data of each monitoring point are used as the node features of the corresponding node.

[0037] The node representation is updated by aggregating the features of each node's neighbor nodes using a graph neural network.

[0038] Calculate the feature vector centrality score for each node based on the updated node representation;

[0039] Nodes whose feature vector centrality scores are consistently higher than their neighboring nodes and exceed the centrality threshold are identified as risk convergence nodes.

[0040] Furthermore, calculating the eigenvector centrality score of each node based on the updated node representation includes: constructing an adjacency matrix between nodes based on the updated node representation; calculating the principal eigenvector of the adjacency matrix using a power iteration algorithm; and using the corresponding element values ​​in the principal eigenvector as the eigenvector centrality score of each node.

[0041] Furthermore, the stress data at risk convergence nodes are identified as the core prediction targets. The extreme values ​​of shrinkage stress and their occurrence times within a set future time period are predicted, and maintenance control commands for these risk convergence nodes are generated and output to the on-site maintenance equipment, including:

[0042] Establish a stress time series based on historical stress data from risk convergence nodes;

[0043] A time series forecasting algorithm is used to predict the stress change trend of risk convergence nodes within a future set time period;

[0044] Extract the extreme values ​​of contractile stress and their corresponding occurrence times from the predicted stress change trends;

[0045] The predicted extreme shrinkage stress was compared with the tensile strength threshold of coarse aggregate UHPC material;

[0046] When the predicted extreme value of shrinkage stress exceeds the tensile strength threshold of the material, a maintenance control instruction containing instructions to increase the intensity of heat preservation and moisture retention is generated.

[0047] The maintenance control commands are transmitted to the on-site maintenance equipment corresponding to the risk convergence node.

[0048] Furthermore, the time series prediction algorithm is used to predict the stress change trend of the risk convergence node within a future set period, including: establishing a time series prediction model based on the historical stress data of the risk convergence node using the exponential smoothing method; using the time series prediction model to predict the stress value sequence within the future set period; and extracting the stress change trend from the predicted stress value sequence.

[0049] On the other hand, the present invention provides a real-time prediction system for shrinkage stress in wet joints of coarse aggregate UHPC, comprising the following modules:

[0050] The data acquisition module is used to acquire stress data, temperature data, and humidity data at the wet joint monitoring points in real time.

[0051] The event recognition module is used to identify temperature abrupt events where the rate of change of ambient temperature exceeds the threshold of normal construction fluctuations, based on temperature data.

[0052] The lag calculation module is used to analyze stress data within a preset time window after a temperature abrupt event occurs, and calculate the response lag time of the stress change rate relative to the temperature abrupt event.

[0053] The stage determination module is used to analyze the correlation characteristics between the stress change rate and the temperature change rate based on stress data, temperature data, and response lag time. When the ratio of the stress change rate to the temperature change rate exceeds a set threshold, it is determined that the stress nonlinear change stage has been entered.

[0054] The risk identification module is used to construct a spatiotemporal map based on stress, temperature and humidity data from all monitoring points during the nonlinear stress change stage. It uses graph neural networks to analyze the spatial dependencies between nodes and identifies risk convergence nodes through dynamic node centrality calculation.

[0055] The instruction generation module is used to identify the stress data of the risk convergence node as the core prediction object, predict the extreme value of shrinkage stress and its occurrence time within a set period in the future, and generate maintenance control instructions for the risk convergence node to be output to the on-site maintenance equipment.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] 1. By establishing a complete computational chain from environmental abrupt change perception to stress risk prediction, real-time dynamic prediction of shrinkage stress in wet joints of coarse aggregate UHPC was achieved. Compared with traditional post-processing analysis methods, by identifying temperature abrupt change events and stress response hysteresis characteristics in real time, feature extraction and state determination are completed the instant the data is generated, effectively eliminating the prediction delay caused by data accumulation in traditional methods. In particular, by constructing a dynamic correlation model between stress change rate and temperature change rate, the critical state of material transition from linear to nonlinear change can be accurately captured, providing accurate time-series judgment basis for subsequent analysis.

[0058] 2. By introducing spatiotemporal graph structure modeling and graph neural network analysis techniques, scattered monitoring points are transformed into a spatially interconnected network of nodes. Dynamic node centrality calculations accurately identify risk convergence areas. This graph-based analysis method not only reflects the spatial dependencies between monitoring points but also captures stress propagation paths through dynamic updates of node features, thereby achieving precise location of key areas. This significantly improves the comprehensiveness and reliability of risk identification. Furthermore, by focusing on the predictive resource allocation for risk convergence nodes, the computational load is greatly reduced while maintaining prediction accuracy. Attached Figure Description

[0059] Figure 1 This is a flowchart of a method for real-time prediction of shrinkage stress in wet joints of coarse aggregate UHPC according to the present invention.

[0060] Figure 2 This is a schematic diagram of the structure of a real-time prediction system for shrinkage stress in wet joints of coarse aggregate UHPC according to the present invention. Detailed Implementation

[0061] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0062] Example 1: Figure 1 This invention presents a method for real-time prediction of shrinkage stress in wet joints of coarse aggregate UHPC, which includes the following steps:

[0063] S1. Real-time acquisition of stress, temperature and humidity data at wet joint monitoring points;

[0064] S2. Based on temperature data, identify temperature abrupt events where the rate of change in ambient temperature exceeds the threshold of normal construction fluctuations.

[0065] S3. For temperature mutation events, analyze the stress data within a preset time window after the event occurs, and calculate the response lag time of the stress change rate relative to the temperature mutation event.

[0066] S4. Based on stress data, temperature data, and response lag time, analyze the correlation characteristics between stress change rate and temperature change rate. When the ratio of stress change rate to temperature change rate exceeds a set threshold, it is determined that the stress nonlinear change stage has been entered.

[0067] S5. During the nonlinear stress change stage, a spatiotemporal map is constructed based on stress data, temperature data and humidity data of all monitoring points. A graph neural network is used to analyze the spatial dependence between nodes, and risk convergence nodes are identified through dynamic node centrality calculation.

[0068] S6. Determine the stress data of the risk convergence node as the core prediction object, predict the extreme value of shrinkage stress and its occurrence time within a set period in the future, and generate maintenance control instructions for the risk convergence node and output them to the on-site maintenance equipment.

[0069] S1. Real-time acquisition of stress, temperature, and humidity data at wet joint monitoring points, specifically implemented as follows:

[0070] When deploying stress sensors, temperature sensors, and humidity sensors on and within key areas of wet joints, it is first necessary to identify the key areas of the wet joint. These key areas include the mid-span location of the wet joint, the area near the supports, and stress concentration areas, such as the connection between the joint and the main beam and the joint ends. The method for identifying key areas is based on bridge design drawings and finite element analysis results, identifying the areas with the most significant stress response under load and temperature changes. Resistance strain gauge sensors are selected for stress sensors. Their working principle is to obtain stress data by measuring the concrete strain and converting it into an electrical signal. During installation, epoxy resin adhesive is used to firmly attach the sensor to the concrete surface, ensuring full contact between the sensor and the concrete surface to avoid measurement errors. Thermocouple sensors are selected for temperature sensors. Their working principle is based on the Seebeck effect to measure temperature changes. During installation, the sensor is embedded approximately 5 cm deep into the concrete to capture the internal temperature distribution. Capacitive humidity sensors are selected for humidity sensors. Their working principle is to measure ambient humidity by detecting changes in the dielectric constant. During installation, the sensor is placed on the surface of the wet joint and a protective cover is added to prevent direct impact from moisture. The spacing between all sensors is determined based on the size of the wet joint and the required monitoring accuracy. For example, in a wet joint longer than 10 meters, the sensor spacing is set to 2 meters to cover the entire joint area. After installation, the sensors must be calibrated. The calibration process involves adjusting the sensor output under known stress, temperature, and humidity conditions to ensure data accuracy.

[0071] When continuously acquiring stress data using stress sensors, temperature data using temperature sensors, and humidity data using humidity sensors, "continuous acquisition" means recording the sensor outputs without interruption at fixed time intervals. The acquisition frequency is set according to the concrete hardening stage and the rate of environmental change. In the early hardening stage after concrete pouring, the acquisition frequency is set to once per minute to capture rapidly changing stress responses; in the later hardening stage, the acquisition frequency is reduced to once every 5 minutes to balance data volume and real-time requirements. The stress data acquisition process includes reading the voltage signal output by the resistance strain gauge, converting the voltage into a strain value using a Wheatstone bridge circuit, and then calculating the stress value from the strain value according to Hooke's Law and the elastic modulus of concrete. The elastic modulus of concrete is obtained through laboratory tests; for example, for C50 concrete, the elastic modulus is taken as 34.5 GPa. The temperature data acquisition process includes reading the millivolt signal output by the thermocouple, converting the signal into a temperature value using cold junction compensation and linearization, with the unit set to degrees Celsius. The humidity data acquisition process includes reading the capacitance value output by the capacitive sensor, converting the capacitance value into a relative humidity value using a calibration curve, with the unit set to percentage. During the data acquisition process, all data is stored digitally and timestamped to ensure time sequence consistency. If a sensor malfunctions, such as an output value exceeding a reasonable range, the acquisition process will automatically trigger a retry mechanism, for example, re-acquiring data within 3 seconds to eliminate transient interference.

[0072] When synchronously transmitting collected stress, temperature, and humidity data to the data storage location, synchronous transmission means that the stress, temperature, and humidity data are packaged and sent at the same time, ensuring time alignment between the data. The transmission protocol uses a TCP / IP-based communication protocol, and the data is packaged in JSON format, containing the sensor number, data value, timestamp, and checksum. The data storage location is set to a cloud database or a local server, the specific choice depending on the on-site network conditions; if the network is stable, data is preferentially transmitted to the cloud database; if the network is interrupted, the data is temporarily stored in a local buffer and automatically retransmitted after the network is restored. The transmission process includes a data encryption step, using the AES encryption algorithm to encrypt data packets to prevent unauthorized access. The transmission frequency is consistent with the acquisition frequency, for example, transmitting data packets once per minute. The data table structure at the data storage location includes fields such as sensor type, data value, unit, timestamp, and status flag, where the status flag indicates whether the data has been verified. During storage, the data is automatically backed up to a redundant storage system, for example, once a day, to prevent data loss. After data storage, it can be retrieved in real time through a query interface for subsequent analysis. The entire transmission and storage process is monitored through logging, which includes transmission time, data volume, and error information to facilitate troubleshooting.

[0073] S2. Based on temperature data, identify temperature abrupt events where the rate of change in ambient temperature exceeds the threshold for fluctuations during normal construction. Specifically, this is implemented as follows:

[0074] When building a time series of temperature data based on real-time acquired temperature data, the real-time acquired temperature data comes from the continuous output of temperature sensors deployed at wet joint monitoring points. This data is recorded at fixed time intervals, such as one temperature value per minute, along with a precise timestamp. The process of building the time series involves arranging the acquired temperature values ​​in chronological order to form an ordered set of data points, where each data point contains a temperature value and its corresponding acquisition time. A sliding window mechanism is used to construct the time series. The window size is set according to the hardening characteristics of concrete and environmental monitoring requirements; for example, a window length of 30 minutes is set to cover sufficient data points for subsequent analysis. Data preprocessing steps include removing outliers. For example, when a temperature value exceeds a reasonable range, such as below -10 degrees Celsius or above 60 degrees Celsius, the data point is marked as invalid and filled using linear interpolation to ensure the continuity of the time series. The time series is stored in an array or list structure for easy subsequent calculation and access. After the time series is built, the data is updated in real time. Newly acquired temperature values ​​are appended to the end of the series, while the oldest data points within the window are removed to maintain a fixed series length. The entire process is implemented through programming, such as using list structures in the Python language to manage and manipulate time series data.

[0075] When calculating the rate of temperature change between adjacent time points in a time series of temperature data, an adjacent time point is defined as the interval between two consecutive data points in the time series. For example, if there is one data point per minute, then the interval between adjacent time points is one minute. The rate of temperature change is calculated using the difference method. Specifically, the temperature difference between adjacent time points is divided by the time interval to obtain the temperature change per unit time, with the unit set to degrees Celsius per minute. During the calculation, the rate of change cannot be calculated for the first data point in the time series, so the calculation starts from the second data point and proceeds point by point. For example, for time points T1 and T2, with temperature values ​​V1 and V2 respectively, the rate of temperature change is (V2-V1) / (T2-T1). The time interval must be consistent with the data acquisition frequency to ensure calculation accuracy. The calculated rate of temperature change also forms a new time series, with each value corresponding to a time point in the original time series. If the data acquisition frequency changes, for example, from once per minute to once every 30 seconds, the time interval is adjusted accordingly to maintain calculation consistency. During the calculation, the rate of temperature change may be positive or negative, representing an upward or downward trend in temperature, respectively. All calculation results will be stored in a new array and time-aligned with the original temperature time series for use in subsequent steps.

[0076] When determining the conventional construction fluctuation threshold based on pre-collected historical construction environment temperature data, the pre-collected historical construction environment temperature data refers to temperature datasets collected from similar engineering projects or the early construction phases of the same bridge. This data covers various weather conditions and construction periods, including daytime and nighttime temperature records for spring, summer, and autumn. The statistical process first calculates the time series of temperature change rates from the historical temperature data, using the same method as described above: dividing the temperature difference between adjacent time points in the historical data by the time interval. Then, a statistical analysis is performed on the historical temperature change rate series, including calculating the mean and standard deviation. The mean reflects the typical level of temperature change, and the standard deviation reflects the dispersion of fluctuations. The conventional construction fluctuation threshold is set based on the mean plus twice the standard deviation to cover the temperature fluctuation range under normal construction conditions. For example, if the average historical temperature change rate is 0.5 degrees Celsius per minute and the standard deviation is 0.2 degrees Celsius per minute, then the conventional construction fluctuation threshold is set to 0.9 degrees Celsius per minute. The threshold setting also needs to consider seasonal factors; for example, in summer, the threshold may be appropriately increased to cope with higher temperature fluctuations. During the statistical process, historical data needs to be cleaned to remove outliers from extreme weather events, such as records during heavy rain or cold waves, to ensure that the threshold represents normal construction conditions. The final normal construction fluctuation threshold is stored as a fixed parameter for real-time comparison and is updated periodically during construction, such as recalculating weekly, to adapt to environmental changes.

[0077] When comparing the calculated rate of temperature change with a conventional construction fluctuation threshold in real time, real-time comparison means comparing the calculated rate of temperature change with the threshold immediately after each new rate of temperature change is calculated. The comparison process uses conditional logic: if the current rate of temperature change is greater than the conventional construction fluctuation threshold, it is marked as a potential abnormal event; otherwise, it is considered a normal fluctuation. The comparison result is output as a Boolean value, such as true indicating that the threshold is exceeded, and false indicating that it is not exceeded. The implementation of real-time comparison relies on a loop checking mechanism; for example, a comparison operation is triggered every time a new rate of temperature change is calculated. During the comparison process, both the rate of temperature change and the threshold use the same unit, such as degrees Celsius per minute, to ensure consistency. If the data acquisition frequency is high, such as once per second, the comparison frequency is increased accordingly to maintain real-time performance. The comparison result is recorded in an event log, including a timestamp, the rate of temperature change value, the threshold, and the comparison result, for easy subsequent analysis and traceability. In addition, the comparison process includes a data verification step, such as checking whether the rate of temperature change is within a reasonable range to avoid misjudgments caused by sensor malfunctions. The entire comparison process is executed automatically through scripts or embedded programs without manual intervention.

[0078] When the rate of temperature change consistently exceeds the threshold for normal construction fluctuations, it is identified as a temperature mutation event. "Consistently exceeding" means that the rate of temperature change exceeds the threshold at multiple consecutive time points, such as five consecutive time points or longer. The specific duration is set according to the characteristics of the concrete material; for example, for concrete in the early hardening stage, the duration is set to 3 minutes to capture rapid stress response. The identification process includes monitoring and comparing the time series of results. If multiple consecutive comparison results are true, a temperature mutation event is triggered. After event identification, the time of occurrence, duration, and maximum rate of temperature change value are recorded, and an event identifier is generated. For example, if the rate of temperature change exceeds the threshold from time point T1 to T5, then T1 is identified as the event start time. The identification logic also includes a de-jittering mechanism to avoid misjudging instantaneous fluctuations as events; for example, requiring that the number of consecutive points exceeding the threshold is not less than a set value. After event identification, the information is passed to subsequent steps for stress analysis, and an alarm is triggered to notify on-site personnel. Event data is stored in a database, including event type, occurrence time, and related parameters, facilitating historical queries and optimization of threshold settings. The entire identification process ensures real-time performance and accuracy to support timely intervention.

[0079] S3. For temperature abrupt events, analyze the stress data within a preset time window after the event occurs, and calculate the response lag time of the stress change rate relative to the temperature abrupt event. Specifically, the implementation is as follows:

[0080] When determining the preset time window starting from the moment of the temperature mutation event, the moment of the temperature mutation event is derived from the temperature mutation event record identified in step S2, which includes the event start timestamp. The preset time window is set based on the stress response characteristics of concrete materials after a temperature mutation. Typically, stress changes are most significant for a period after the event. Therefore, the length of the preset time window is adjusted according to the concrete type and environmental conditions. For example, for ordinary silicate concrete, the preset time window is set to 30 minutes to cover the entire process from initial stress response to stability. The starting point of the preset time window strictly corresponds to the moment of the temperature mutation event, and the ending point is the starting point plus the window length. For example, if the event occurs at 10:00, the window range is from 10:00 to 10:30. The determination of the window length also references laboratory test data. For example, by simulating concrete stress tests under temperature mutation, the time range required for stress changes to reach their peak is observed, thus setting the window length between 20 and 40 minutes. In practical applications, the preset time window is implemented through program code, for example, using a time calculation function to generate the window boundary based on the moment of the event, plus a fixed interval. Window parameters are stored in a configuration file, allowing for dynamic adjustment based on site conditions. For example, during winter construction, the window length may be extended to 40 minutes due to the slower hardening of concrete. The entire process ensures precise alignment of the time window with temperature abrupt events, providing an accurate time range for subsequent stress analysis.

[0081] When extracting stress data and calculating the stress change rate within a preset time window, the stress data extraction process is based on the real-time stress data acquired in step S1. This data is stored in time series form, with each data point containing a stress value and a corresponding timestamp. During extraction, based on the start and end times of the preset time window, all data points whose timestamps fall within the window are selected from the stress data sequence to form a stress subsequence within the window. For example, if the preset time window is from 10:00 to 10:30, then all stress values ​​with timestamps within this interval are extracted. The method for calculating the stress change rate is differential calculation, that is, taking the stress difference between adjacent time points in the stress subsequence and dividing it by the time interval to obtain the stress change per unit time, with the unit set to megapascals per minute. The definition of adjacent time points is consistent with the data acquisition frequency; for example, if the data acquisition frequency is once per minute, then the time interval is 1 minute. During calculation, calculation is performed point by point starting from the second data point in the stress subsequence. For example, for the stress values ​​K1 and K2 at time points T1 and T2, the stress change rate is (K2-K1) / (T2-T1). The calculation results form a new stress change rate sequence, with each value corresponding to a time point in the original stress subsequence. If data is missing, for example due to temporary sensor malfunction, linear interpolation is used to fill in the stress values ​​to ensure computational continuity. All calculated stress change rate values ​​are stored in an array and associated with timestamps for subsequent trend analysis.

[0082] When analyzing the trend of stress change rate within a preset time window and identifying the inflection point from the initial response to significant acceleration, the trend analysis is based on the stress change rate sequence. Patterns are identified by observing the overall trend of the sequence values. The initial response stage corresponds to a relatively flat stress change rate range, the significant acceleration stage corresponds to a range where the stress change rate rises rapidly, and the inflection point is the turning point between these two stages. The method for identifying the inflection point uses the numerical difference method. Specifically, the first-order difference sequence of the stress change rate sequence is calculated, which is the difference between adjacent stress change rate values. Then, the maximum value point is found in the first-order difference sequence; this point corresponds to the moment when the stress change rate accelerates most significantly. For example, if the stress change rate sequence is [0.1, 0.2, 0.5, 1.0] MPa per minute, then the first-order difference sequence is [0.1, 0.3, 0.5], where the position of the original sequence corresponding to the maximum value of 0.5 is the inflection point. Inflection point identification also requires threshold judgment. For example, the first-order difference value must exceed a preset acceleration threshold. This threshold is set based on historical stress data statistics, such as taking the average of historical stress change rate difference values ​​plus one standard deviation to avoid noise interference. The identification process is achieved by iteratively traversing the first-order difference sequence, finding the first point that meets the conditions as the inflection point, and recording its corresponding timestamp. If there are multiple candidate points in the sequence, the point with the largest difference value is selected to ensure that the inflection point represents the most significant acceleration change. After inflection point identification, its timestamp and stress change rate value are stored in the event log for subsequent lag time calculation.

[0083] The time difference between the occurrence of a temperature abrupt change event and the occurrence of the inflection point of the stress change rate is calculated. This time difference is used as the response lag time of the stress change rate relative to the temperature abrupt change event. The occurrence time of the temperature abrupt change event is derived from the event start timestamp recorded in step S2, and the occurrence time of the inflection point of the stress change rate is derived from the inflection point timestamp identified in the preceding steps. The time difference is calculated using simple subtraction: the occurrence time of the temperature abrupt change event is subtracted from the occurrence time of the inflection point, resulting in a difference in minutes. For example, if the temperature abrupt change event occurs at 10:00 and the inflection point occurs at 10:12, the time difference is 12 minutes. During the calculation, timestamps are consistently used in the same time format, such as Unix timestamps or ISO 8601 format, to ensure calculation accuracy. The response lag time represents the delay time during which a temperature change significantly accelerates the stress response. This value is affected by the concrete material properties and environmental conditions; for example, the response lag time of early-hardening concrete is usually shorter. The calculated response lag time is stored in a database with an event identifier added for easy use in subsequent analysis. If the inflection point is not identified within the preset time window, the response lag time is marked as invalid, triggering a re-analysis mechanism. The entire calculation process is executed automatically via script, and the results are used to optimize the stress prediction model, such as adjusting the analysis time parameters in subsequent steps. The response lag time data can also be used for statistical reporting to help evaluate the behavior of concrete under different temperature conditions.

[0084] S4. Based on stress data, temperature data, and response lag time, analyze the correlation characteristics between the stress change rate and the temperature change rate. When the ratio of the stress change rate to the temperature change rate exceeds a set threshold, it is determined that the stress has entered the nonlinear change stage. Specifically, this is implemented as follows:

[0085] When determining the corresponding analysis period for stress and temperature data based on the response lag time, the response lag time is derived from the response lag time value of the stress change rate relative to the temperature abrupt event calculated in step S3. This value represents the delay time for a significant stress response triggered by a temperature change. The method for determining the corresponding analysis period is to use the time of the temperature abrupt event as a baseline, add the response lag time as the starting point of the analysis period, and then set a fixed duration as the duration of the analysis period, for example, 30 minutes, thus forming a time interval from the starting point to the starting point plus 30 minutes. An example of calculating the starting point of the analysis period is that if the temperature abrupt event occurs at 10:00 and the response lag time is 5 minutes, then the analysis period starts at 10:05 and ends at 10:35. The duration of the analysis period is adjusted according to the stress response characteristics of concrete. For example, for coarse aggregate UHPC materials, because the stress change is slower in the early hardening stage, the duration may be extended to 40 minutes to ensure complete coverage of the stress-temperature interaction process. The corresponding analysis period is implemented automatically through program code. For example, a time calculation function is used, taking the time of the temperature abrupt event and the response lag time as input, to generate the start and end timestamps of the analysis period. The parameters of the analysis period are stored in a configuration table, facilitating dynamic optimization based on field monitoring data. For instance, the duration can be adjusted weekly based on newly acquired response lag time statistics. The entire determination process ensures that stress data and temperature data are precisely aligned in time, providing a foundation for subsequent correlation analysis.

[0086] When calculating the rate of stress change and the rate of temperature change within the corresponding analysis period, the stress data is derived from the real-time stress data sequence acquired in step S1, and the temperature data is derived from the real-time temperature data sequence acquired in step S1. The corresponding analysis period is defined by the start and end time ranges determined in the preceding steps. The method for calculating the rate of stress change is as follows: extract all data points within the analysis period from the stress data sequence, then calculate the difference between stress values ​​at adjacent time points and divide by the time interval to obtain the stress change per unit time, with the unit set to megapascals per minute. The method for calculating the rate of temperature change is similar: extract data points within the same analysis period from the temperature data sequence, calculate the difference between temperature values ​​at adjacent time points and divide by the time interval to obtain the temperature change per unit time, with the unit set to degrees Celsius per minute. The definition of adjacent time points is consistent with the data acquisition frequency; for example, if the data acquisition frequency is once per minute, the time interval is 1 minute. During the calculation process, if data points are missing, linear interpolation is used to fill them in to ensure sequence continuity. The calculation results of the rate of stress change and the rate of temperature change form two independent time series, with the values ​​in each series corresponding to each time point within the analysis period. All calculations are performed automatically by looping through the data points within the analysis period, and the results are stored in an array for subsequent ratio calculations.

[0087] When calculating the real-time ratio of the stress change rate to the temperature change rate, the real-time ratio refers to the quotient obtained by dividing the stress change rate value by the temperature change rate value at each time point. This ratio is dimensionless and reflects the sensitivity of stress change to temperature change. The calculation process is based on the stress change rate sequence and temperature change rate sequence obtained in the preceding steps, requiring that the two sequences be perfectly aligned in time, meaning that each time point has a corresponding stress change rate value and temperature change rate value. For each time point, the real-time ratio equals the stress change rate value divided by the temperature change rate value. For example, if the stress change rate is 0.5 MPa per minute and the temperature change rate is 0.2 degrees Celsius per minute at a certain time point, then the real-time ratio is 2.5. During calculation, if the temperature change rate value is zero, the point is skipped to avoid division by zero errors and marked as invalid data. The calculation frequency of the real-time ratio is consistent with the data acquisition frequency, for example, once per minute, forming a new ratio sequence. The calculation results are updated in real time and stored in a database, with timestamps added for tracking. The ratio sequence is used for subsequent threshold comparisons. The entire process is implemented through scripts to ensure efficiency and accuracy.

[0088] When setting the stress-temperature coupled response threshold based on the material properties of coarse aggregate UHPC, these properties include parameters such as tensile strength, elastic modulus, and shrinkage coefficient. These parameters are obtained through laboratory tests; for example, tests on standard concrete specimens yielded a tensile strength of 5 MPa and an elastic modulus of 40 GPa. The stress-temperature coupled response threshold is set based on the material's sensitivity to temperature changes. Specifically, a critical ratio is determined through historical data statistics or empirical formulas. When the ratio of the stress change rate to the temperature change rate exceeds this critical value, the material enters the nonlinear response stage. For example, based on experimental data of coarse aggregate UHPC, the threshold might be set to 2.0, indicating that a nonlinear response begins when the stress change rate reaches twice the temperature change rate. Environmental factors must also be considered when setting the threshold. For instance, under high-temperature conditions, the threshold might be appropriately lowered to 1.8 to accommodate faster stress accumulation. The threshold is stored as a fixed parameter in a configuration file and can be revised periodically based on field monitoring results, such as monthly reassessment. The setup process includes data validation, such as checking whether the threshold is within a reasonable range, such as between 1.5 and 3.0, to avoid misjudgments. The entire setup ensures that the threshold matches the actual behavior of the material, supporting accurate state determination.

[0089] When the real-time ratio of the stress change rate to the temperature change rate consistently exceeds the stress-temperature coupling response threshold, the system is determined to have entered the stress nonlinear change stage. "Consistently exceeding" means that the real-time ratio is greater than the stress-temperature coupling response threshold at multiple consecutive time points, such as five consecutive time points or longer. The specific number of time points is set according to the concrete hardening rate; for example, for the early hardening stage, the number of time points is set to 3 minutes. The determination process involves real-time monitoring of the ratio sequence. If, starting from a certain time point, multiple consecutive ratio values ​​exceed the threshold, then that time point is marked as entering the stress nonlinear change stage. For example, if the stress-temperature coupling response threshold is 2.0, and the ratio sequence is [2.1, 2.3, 2.5, 2.4, 2.6], and five consecutive points exceed the threshold, then the system is determined to have entered the nonlinear stage at the time corresponding to the first point. The determination logic is implemented using a counter. For example, the counter is initialized to zero; when the ratio exceeds the threshold, the counter is incremented by one; otherwise, it is reset to zero. When the counter reaches a set number of points, the determination is triggered. The determination result includes the timestamp of entering the stage and the duration, and is recorded in the event log. If the ratio falls below the threshold, the stage state remains unchanged until a new decision is triggered. The entire decision process is executed automatically and triggers subsequent analysis steps, such as the spatiotemporal graph construction in step S5. The decision data is used to optimize the threshold and model parameters, improving prediction accuracy.

[0090] S5. During the nonlinear stress change stage, a spatiotemporal map is constructed based on stress, temperature, and humidity data from all monitoring points. A graph neural network is used to analyze the spatial dependencies between nodes, and risk convergence nodes are identified through dynamic node centrality calculation. Specifically, the implementation is as follows:

[0091] During the nonlinear stress change phase, when constructing the spatiotemporal map based on stress, temperature, and humidity data from all monitoring points, the nonlinear stress change phase originates from the entry state determined in step S4. All monitoring points originate from the sensor locations arranged in step S1, and the stress, temperature, and humidity data originate from the continuously collected data sequence in step S1. The spatiotemporal map construction process uses each monitoring point as a graph node, with node identifiers corresponding one-to-one with monitoring point numbers. For example, monitoring point A1 corresponds to node N1, and monitoring point A2 corresponds to node N2, ensuring that nodes are associated with their physical locations. Edges in the graph structure are defined based on the spatial distance between monitoring points. The spatial distance is calculated using the coordinates of the monitoring points, which are derived from construction drawings or on-site measurements, such as using a total station to obtain the three-dimensional coordinates of each monitoring point. The rule for establishing edges is that if the Euclidean distance between two monitoring points is less than a preset distance threshold, an edge is added between these two corresponding nodes. The preset distance threshold is set according to the wet joint size and monitoring density; for example, for a 20-meter-long wet joint, the threshold is set to 5 meters to capture local interactions. The graph structure is stored as an adjacency list, where each node records a list of its neighboring nodes, facilitating subsequent processing by the graph neural network. The construction process also includes setting edge weights, with weights inversely proportional to spatial distance; for example, smaller distances result in higher weights, reflecting the strength of neighborly influence. The entire graph structure is updated every 5 minutes to incorporate the latest monitoring data, ensuring that the spatiotemporal graph dynamically reflects on-site conditions.

[0092] When stress, temperature, and humidity data from each monitoring point are used as node features for the corresponding node, the node feature is defined as a feature vector associated with each node, which consists of the real-time data values ​​of the monitoring point. The feature vector has a fixed dimension of 3, corresponding to stress, temperature, and humidity data respectively. For example, for node N1, the feature vector is [stress value, temperature value, humidity value]. The data values ​​come from the real-time sequence collected in step S1, taking the data from the latest timestamp of each monitoring point. For example, if the current time is 10:00, then the stress, temperature, and humidity readings at 10:00 are used. The feature data is normalized before input to avoid the influence of dimensional differences. The normalization method uses min-max scaling, scaling each feature value to between 0 and 1. For example, if the original stress value range is 0 to 10 MPa, the normalized value is the original value divided by 10. The node features are updated synchronously with the data acquisition frequency, for example, once per minute, to ensure that the feature vector reflects the latest state. The features are stored in the node attribute table and combined with the graph structure to form a complete graph data input. For missing data, the nearest neighbor interpolation method is used to fill in the gaps, for example, by using the last valid reading at the same monitoring point to replace it, in order to ensure the continuity of characteristics.

[0093] This paper utilizes a graph neural network (Graph Neural Network) to aggregate the features of each node's neighbors. When updating the node representation, the Graph Neural Network employs a two-layer graph convolutional network: the first layer is used for feature transformation, and the second layer is used for feature aggregation. The aggregation process is based on the adjacency relationships of the graph. For each node, the feature vectors of all its neighbors are first collected. Then, the neighbor features are aggregated using a weighted summation method. The weights are derived from the edge weights; for example, if the edge weight between nodes N1 and N2 is 0.8, then the contribution of N2's feature to N1 is multiplied by 0.8. The aggregated neighbor features are concatenated with the node's own features to form an extended feature vector. This extended feature vector is then transformed non-linearly through a fully connected layer and an activation function to generate the updated node representation. The ReLU activation function is chosen to introduce non-linearity. The update process iterates only once; each node only aggregates the features of its direct neighbors, without multi-hop propagation, to maintain computational efficiency. The dimension of the node representation is consistent with the dimension of the input features; for example, a 3D input remains a 3D representation after transformation, ensuring that the output scale is compatible with the input. The entire graph neural network operates using pre-trained parameters derived from a model trained on historical data. The model is updated quarterly to adapt to changes in materials. The updated node representations are stored in a new feature matrix for subsequent centrality calculations.

[0094] When calculating the eigenvector centrality score of each node based on the updated node representation, an adjacency matrix is ​​first constructed based on the updated node representation. The size of the adjacency matrix is ​​the number of nodes multiplied by the number of nodes. The matrix element values ​​represent the similarity between nodes, which is obtained by calculating the cosine similarity between node representations. For example, for nodes N1 and N2, whose representation vectors are V1 and V2, the similarity is equal to the dot product of V1 and V2 divided by the product of the magnitudes of V1 and V2. After the adjacency matrix is ​​constructed, its principal eigenvector is calculated using a power iteration algorithm. The power iteration algorithm involves initializing a random vector as an eigenvector estimate, then repeatedly multiplying the adjacency matrix by this vector and normalizing the result until the vector change is less than a convergence threshold, for example, set to 0.001. The final stable vector is the principal eigenvector. Each element value in the principal eigenvector corresponds to the eigenvector centrality score of a node, with the score ranging from 0 to 1. A higher value indicates a greater influence of the node in the graph. The calculation process is executed every 10 minutes to reflect the dynamic changes in node representations. The eigenvector centrality score is stored in association with the node identifier, facilitating subsequent risk identification. If the adjacency matrix is ​​sparse, a sparse optimization algorithm is used to improve efficiency.

[0095] When identifying nodes whose eigenvector centrality scores are consistently higher than their neighboring nodes and exceed the centrality threshold as risk convergence nodes, "consistently higher" means that the node's eigenvector centrality score is higher than the scores of all its neighboring nodes for multiple consecutive calculation periods, such as three consecutive periods, each period being 10 minutes. The centrality threshold is set based on historical centrality score statistics, for example, taking the median of all node scores plus one standard deviation, and the threshold is updated quarterly. The identification process involves comparing each node's current score with the list of neighboring node scores. If the node's score is greater than the scores of all neighboring nodes and also greater than the centrality threshold, it is marked as a candidate risk node. Then, it is checked whether the candidate node maintains this condition in subsequent periods. If it continuously meets the condition, it is finally identified as a risk convergence node. The identification results are recorded in the event log, including the node number, identification time, and relevant score value. For boundary cases, such as when scores are equal, the node with the higher score is selected first; if they are still equal, a node is selected randomly. The identification process is executed every 10 minutes to ensure real-time performance. The data from risk convergence nodes is used for prediction and intervention in subsequent steps, and also triggers alarms to notify on-site personnel. The entire recognition logic is implemented automatically through scripts, and the accuracy is evaluated regularly to optimize the threshold parameters.

[0096] S6. The stress data at risk convergence nodes is identified as the core prediction target. The extreme values ​​of shrinkage stress and their occurrence times within a set future time period are predicted. Maintenance control commands for these risk convergence nodes are then generated and output to the on-site maintenance equipment. The specific implementation is as follows:

[0097] When constructing a stress time series based on historical stress data from risk convergence nodes, the risk convergence nodes are derived from the node set identified in step S5, and the historical stress data are derived from the stress data sequences continuously collected for these nodes in step S1. The data covers a fixed time period from the current moment back, such as all stress readings within 24 hours. The process of constructing the stress time series includes extracting stress data for each risk convergence node from the storage database, arranging them in ascending order by timestamp to form an ordered list of data points, each data point containing a stress value and the corresponding collection time. The time series is constructed using a sliding window mechanism, with the window size set according to the prediction requirements, for example, a window length of 12 hours, to include sufficient historical information for trend analysis. Data preprocessing steps include removing outliers; for example, when the stress value exceeds a reasonable range, such as below 0 MPa or above 10 MPa, the data point is marked as invalid and filled using linear interpolation to ensure the continuity and integrity of the time series. The time series is stored in an array structure, with each risk convergence node corresponding to an independent sequence. The sequence data is updated in real time, with newly acquired stress values ​​appended to the end of the sequence while the oldest data points within the window are removed to maintain a fixed sequence length. The entire process is automated through database queries and data processing scripts, ensuring that the time series accurately reflects the stress history of the nodes.

[0098] When using time series forecasting algorithms to predict stress change trends within a future set time period for risk convergence nodes, the exponential smoothing method is selected. The length of the future set time period is set according to the maintenance response time, for example, 30 minutes, to provide sufficient early warning time. The forecasting process is based on the stress time series established in the preceding steps. First, an exponential smoothing method is used to establish a time series forecasting model. The core steps of the exponential smoothing method include calculating the smoothing coefficient, which is adjusted according to the volatility of historical data. For example, a coefficient value of 0.3 is selected through trial and error to balance the influence of recent and long-term data. After the model is established, the time series forecasting model is used to predict the stress value sequence within the future set time period. During forecasting, stress values ​​for future time points are generated point by point. For example, starting from the current moment, one stress value is predicted every minute, generating a total of 30 predicted values. The forecast results form a new stress value sequence, representing the stress change trajectory within the future time period. The stress change trend is extracted from the predicted stress value sequence by calculating the first difference of the sequence, i.e., the difference between adjacent predicted stress values, forming a trend sequence. The positive or negative trend value indicates an increase or decrease in stress. The entire prediction process is executed every 10 minutes to incorporate the latest data, and the prediction results are stored in the prediction database for subsequent extreme value extraction.

[0099] When extracting the extreme values ​​of shrinkage stress and their corresponding occurrence times from the predicted stress change trend, the extreme value of shrinkage stress is defined as the maximum value in the predicted stress value sequence, representing the highest possible level of shrinkage stress in the future. The extraction process involves traversing the predicted stress value sequence, finding the maximum stress value in the sequence, and recording the timestamp corresponding to that value as the occurrence time. For example, if the predicted sequence is [3.0, 3.2, 3.5, 3.3] MPa, and the corresponding timestamps are [10:00, 10:01, 10:02, 10:03], then the extreme value of shrinkage stress is 3.5 MPa, and the occurrence time is 10:02. During extraction, if multiple identical maximum values ​​exist in the sequence, the first occurrence timestamp is selected to ensure uniqueness. The extraction process also includes a verification step, such as checking whether the extreme value is within a reasonable range, such as not exceeding the material's theoretical limits, to avoid anomalies caused by prediction errors. The extraction results are stored in an event log, including the extreme value value, occurrence time, and associated node identifier, for subsequent threshold comparison. The entire extraction process is implemented using a loop algorithm, and timestamps are added to ensure traceability.

[0100] When comparing the predicted extreme shrinkage stress with the tensile strength threshold of coarse aggregate UHPC material, the tensile strength threshold is derived from laboratory test data obtained through tensile tests on standard specimens. For example, for C50 strength grade coarse aggregate UHPC, the tensile strength threshold is set to 5 MPa. The comparison process uses numerical comparison, that is, subtracting the tensile strength threshold from the predicted extreme shrinkage stress. If the difference is greater than zero, it indicates that the threshold is exceeded. The unit is consistently set to MPa to ensure dimensional consistency. The comparison logic is implemented through conditional judgment. For example, if the predicted extreme value is 5.2 MPa and the tensile strength threshold is 5.0 MPa, the comparison result is that the threshold is exceeded. The comparison process is executed synchronously with the prediction cycle, for example, every 10 minutes, to ensure real-time performance. The comparison result is output as a Boolean value and recorded in the comparison log, including the comparison time, extreme value, threshold, and result flag. If the tensile strength threshold needs to be adjusted, for example based on changes in environmental humidity, a new value is dynamically loaded from the material parameter library to maintain comparison accuracy.

[0101] When the predicted extreme shrinkage stress exceeds the material's tensile strength threshold, a maintenance control instruction containing instructions to increase the strength of thermal insulation and moisture retention maintenance is generated. The content of the maintenance control instruction is a text string or structured data, specifying specific maintenance actions, such as increasing the coverage density of the thermal insulation blanket or increasing the spraying frequency of the misting system. Instruction generation is based on the aforementioned comparison results; if the comparison result is true, the instruction generation process is automatically triggered. The generation process includes defining instruction parameters, such as setting the increase in thermal insulation strength to 20% and the increase in moisture retention strength to 15%. The parameter values ​​are dynamically adjusted according to the degree of stress exceeding the limit; for example, if the predicted extreme value exceeds the threshold of 0.5 MPa, the increase is increased to 25%. The instruction encoding uses JSON format and includes fields such as instruction type, target node, strength parameter, and execution time, for example, {"instruction": "increase_maintenance", "node": "N1", "insulation_strength":20, "humidity_strength": 15, "time": "10:00"}. Once generated, instructions are stored in an instruction queue with a priority flag added to ensure timely processing. The generation process also includes logging for auditing and optimization.

[0102] When transmitting maintenance control commands to the corresponding field maintenance equipment at the risk convergence node, this equipment includes insulation and humidification devices such as electric blankets and automatic sprayers. These devices are associated with the risk convergence node through unique identifiers. The transmission process uses wireless communication protocols, such as MQTT. Data is encapsulated into data packets containing the command content, the target device address, and a checksum. Before transmission, the data packets are encrypted using the AES algorithm to ensure security. The transmission frequency is synchronized with command generation, for example, once every 10 minutes. If there is network delay, a retry mechanism is initiated, with a maximum of 3 retries. After transmission, the device controller receives, parses, and executes the command, such as adjusting heating power or spray interval. Transmission status monitoring is achieved through an acknowledgment mechanism; the device returns an execution response. If no response is received within a timeout period, the transmission is marked as failed and an alarm is triggered. The entire transmission process ensures accurate delivery of commands and logs the transmission time and results for subsequent analysis.

[0103] Example 2: Figure 2 A schematic diagram of the structure of a real-time prediction system for shrinkage stress in wet joints of coarse aggregate UHPC is provided. The system includes the following modules:

[0104] The data acquisition module is used to acquire stress data, temperature data, and humidity data at the wet joint monitoring points in real time.

[0105] The event recognition module is used to identify temperature abrupt events where the rate of change of ambient temperature exceeds the threshold of normal construction fluctuations, based on temperature data.

[0106] The lag calculation module is used to analyze stress data within a preset time window after a temperature abrupt event occurs, and calculate the response lag time of the stress change rate relative to the temperature abrupt event.

[0107] The stage determination module is used to analyze the correlation characteristics between the stress change rate and the temperature change rate based on stress data, temperature data, and response lag time. When the ratio of the stress change rate to the temperature change rate exceeds a set threshold, it is determined that the stress nonlinear change stage has been entered.

[0108] The risk identification module is used to construct a spatiotemporal map based on stress, temperature and humidity data from all monitoring points during the nonlinear stress change stage. It uses graph neural networks to analyze the spatial dependencies between nodes and identifies risk convergence nodes through dynamic node centrality calculation.

[0109] The instruction generation module is used to identify the stress data of the risk convergence node as the core prediction object, predict the extreme value of shrinkage stress and its occurrence time within a set period in the future, and generate maintenance control instructions for the risk convergence node to be output to the on-site maintenance equipment.

[0110] All calculations involved in the embodiments are dimensionless numerical calculations, and the preset parameters and thresholds in the calculations are set by those skilled in the art according to the actual situation.

[0111] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0112] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and inventive constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0113] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0114] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0115] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0116] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for real-time prediction of shrinkage stress in wet joints of coarse aggregate UHPC, characterized in that, Includes the following steps: S1. Real-time acquisition of stress, temperature and humidity data at wet joint monitoring points; S2. Based on temperature data, identify temperature abrupt events where the rate of change in ambient temperature exceeds the threshold of normal construction fluctuations. S3. For temperature mutation events, analyze the stress data within a preset time window after the event occurs, and calculate the response lag time of the stress change rate relative to the temperature mutation event. S4. Based on stress data, temperature data, and response lag time, analyze the correlation characteristics between stress change rate and temperature change rate. When the ratio of stress change rate to temperature change rate exceeds a set threshold, it is determined that the stress nonlinear change stage has been entered. S5. During the nonlinear stress change stage, a spatiotemporal map is constructed based on stress data, temperature data and humidity data of all monitoring points. A graph neural network is used to analyze the spatial dependence between nodes, and risk convergence nodes are identified through dynamic node centrality calculation. S6. Determine the stress data of the risk convergence node as the core prediction object, predict the extreme value of shrinkage stress and its occurrence time within a set period in the future, and generate maintenance control instructions for the risk convergence node and output them to the on-site maintenance equipment.

2. The method for real-time prediction of shrinkage stress in wet joints of coarse aggregate UHPC according to claim 1, characterized in that, Real-time acquisition of stress, temperature, and humidity data at wet joint monitoring points, including: Stress sensors, temperature sensors, and humidity sensors are installed on the surface and in key areas inside the wet joint. Stress data is continuously collected using a stress sensor, temperature data is continuously collected using a temperature sensor, and humidity data is continuously collected using a humidity sensor. The collected stress, temperature, and humidity data are simultaneously transmitted to the data storage location.

3. The method for real-time prediction of shrinkage stress in wet joints of coarse aggregate UHPC according to claim 1, characterized in that, Based on temperature data, identify temperature abrupt events where the rate of change in ambient temperature exceeds the threshold for typical construction fluctuations, including: Establish a time series of temperature data based on real-time acquired temperature data; Calculate the rate of temperature change between adjacent time points in a time series of temperature data; The threshold for fluctuations in conventional construction conditions is obtained by statistically analyzing historical construction environment temperature data collected in advance. The calculated rate of temperature change is compared in real time with the conventional construction fluctuation threshold. When the rate of temperature change continuously exceeds the normal construction fluctuation threshold, it is identified as a temperature abrupt event.

4. The method for real-time prediction of shrinkage stress in wet joints of coarse aggregate UHPC according to claim 1, characterized in that, For a sudden temperature event, stress data within a preset time window after its occurrence is analyzed, and the response lag time of the stress change rate relative to the sudden temperature event is calculated, including: Determine a preset time window starting from the moment the temperature abrupt change event occurs; Extract stress data and calculate the rate of stress change within a preset time window; Analyze the trend of stress change rate within a preset time window to identify the inflection point from the initial response to a significant acceleration in stress change rate; Calculate the time difference between the occurrence of the temperature abrupt change event and the occurrence of the inflection point of the stress change rate, and use this time difference as the response lag time of the stress change rate relative to the temperature abrupt change event.

5. The method for real-time prediction of shrinkage stress in wet joints of coarse aggregate UHPC according to claim 1, characterized in that, Based on stress data, temperature data, and response hysteresis time, the correlation characteristics between the stress change rate and the temperature change rate are analyzed. When the ratio of the stress change rate to the temperature change rate exceeds a set threshold, it is determined that the stress has entered a nonlinear change stage, including: The corresponding analysis period for stress data and temperature data is determined based on the response lag time. Calculate the rate of stress change and the rate of temperature change within the corresponding analysis period; Calculate the real-time ratio of the rate of stress change to the rate of temperature change; The stress-temperature coupling response threshold is set based on the material properties of coarse aggregate UHPC. When the real-time ratio of the rate of stress change to the rate of temperature change continuously exceeds the stress-temperature coupling response threshold, it is determined that the stress has entered the nonlinear change stage.

6. The method for real-time prediction of shrinkage stress in wet joints of coarse aggregate UHPC according to claim 1, characterized in that, During the nonlinear stress variation phase, a spatiotemporal map is constructed based on stress, temperature, and humidity data from all monitoring points. A graph neural network is used to analyze the spatial dependencies between nodes, and risk convergence nodes are identified through dynamic node centrality calculations, including: Each monitoring point is used as a graph node, and the spatial distance between monitoring points is used as an edge to construct the graph structure; The stress data, temperature data, and humidity data of each monitoring point are used as the node features of the corresponding node. The node representation is updated by aggregating the features of each node's neighbor nodes using a graph neural network. Calculate the feature vector centrality score for each node based on the updated node representation; Nodes whose feature vector centrality scores are consistently higher than their neighboring nodes and exceed the centrality threshold are identified as risk convergence nodes.

7. The method for real-time prediction of shrinkage stress in wet joints of coarse aggregate UHPC according to claim 6, characterized in that, Calculating the eigenvector centrality score of each node based on the updated node representation includes: constructing an adjacency matrix between nodes based on the updated node representation; calculating the principal eigenvector of the adjacency matrix using a power iteration algorithm; and using the corresponding element values ​​in the principal eigenvector as the eigenvector centrality score of each node.

8. The method for real-time prediction of shrinkage stress in wet joints of coarse aggregate UHPC according to claim 1, characterized in that, The stress data at risk convergence points are identified as the core prediction targets. The extreme values ​​of shrinkage stress and their occurrence times within a specified future time period are predicted. Maintenance control commands for these risk convergence points are then generated and output to the on-site maintenance equipment, including: Establish a stress time series based on historical stress data from risk convergence nodes; A time series forecasting algorithm is used to predict the stress change trend of risk convergence nodes within a future set time period; Extract the extreme values ​​of contractile stress and their corresponding occurrence times from the predicted stress change trends; The predicted extreme shrinkage stress was compared with the tensile strength threshold of coarse aggregate UHPC material; When the predicted extreme value of shrinkage stress exceeds the tensile strength threshold of the material, a maintenance control instruction containing instructions to increase the intensity of heat preservation and moisture retention is generated. The maintenance control commands are transmitted to the on-site maintenance equipment corresponding to the risk convergence node.

9. A method for real-time prediction of shrinkage stress in wet joints of coarse aggregate UHPC according to claim 8, characterized in that, The method of using time series forecasting algorithms to predict stress change trends in a future set period for risk convergence nodes includes: establishing a time series forecasting model based on historical stress data of risk convergence nodes using the exponential smoothing method; using the time series forecasting model to predict stress value sequences in the future set period; and extracting stress change trends from the predicted stress value sequences.

10. A real-time prediction system for shrinkage stress in wet joints of coarse aggregate UHPC, used to implement the real-time prediction method for shrinkage stress in wet joints of coarse aggregate UHPC as described in any one of claims 1-9, characterized in that, Includes the following modules: The data acquisition module is used to acquire stress data, temperature data, and humidity data at the wet joint monitoring points in real time. The event recognition module is used to identify temperature abrupt events where the rate of change of ambient temperature exceeds the threshold of normal construction fluctuations, based on temperature data. The lag calculation module is used to analyze stress data within a preset time window after a temperature abrupt event occurs, and calculate the response lag time of the stress change rate relative to the temperature abrupt event. The stage determination module is used to analyze the correlation characteristics between the stress change rate and the temperature change rate based on stress data, temperature data, and response lag time. When the ratio of the stress change rate to the temperature change rate exceeds a set threshold, it is determined that the stress nonlinear change stage has been entered. The risk identification module is used to construct a spatiotemporal map based on stress, temperature and humidity data from all monitoring points during the nonlinear stress change stage. It uses graph neural networks to analyze the spatial dependencies between nodes and identifies risk convergence nodes through dynamic node centrality calculation. The instruction generation module is used to identify the stress data of the risk convergence node as the core prediction object, predict the extreme value of shrinkage stress and its occurrence time within a set period in the future, and generate maintenance control instructions for the risk convergence node to be output to the on-site maintenance equipment.

Images